An anisotropic cohesive fracture model: advantages and ...
Transcript of An anisotropic cohesive fracture model: advantages and ...
An anisotropic cohesive fracture model: advantages and limitations of
length-scale insensitive phase-field damage models
Shahed Rezaei1, Ali Harandi2, Tim Brepols2, Stefanie Reese2
1Mechanics of Functional Materials Division, Institute of Materials Science,Technische Universitat Darmstadt, Darmstadt 64287, Germany
2Institute of Applied Mechanics,RWTH Aachen University, D-52074 Aachen, Germany
Abstract
The goal of the current work is to explore direction-dependent damage initiation and propa-
gation within an arbitrary anisotropic solid. In particular, we aim at developing anisotropic
cohesive phase-field (PF) damage models by extending the idea introduced in [1] for direction-
dependent fracture energy and also anisotropic PF damage models based on structural tensors.
The cohesive PF damage formulation used in the current contribution is motivated by the works
of [2, 3, 4]. The results of the latter models are shown to be insensitive with respect to the
length scale parameter for the isotropic case. This is because they manage to formulate the
fracture energy as a function of diffuse displacement jumps in the localized damaged zone. In
the present paper, we discuss numerical examples and details on finite element implementa-
tions where the fracture energy, as well as the material strength, are introduced as an arbitrary
function of the crack direction. Using the current formulation for anisotropic cohesive fracture,
the obtained results are almost insensitive with respect to the length scale parameter. The
latter is achieved by including the direction-dependent strength of the material in addition to
its fracture energy. Utilizing the current formulation, one can increase the mesh size which
reduces the computational time significantly without any severe change in the predicted crack
path and overall obtained load-displacement curves. We also argue that these models still lack
to capture mode-dependent fracture properties. Open issues and possible remedies for future
developments are finally discussed as well.
Keywords: anisotropic cohesive fracture, phase-field damage model, length-scale insensitive
Preprint submitted to Authors August 31, 2021
arX
iv:2
108.
1267
5v1
[cs
.CE
] 2
8 A
ug 2
021
1. Introduction
Understanding and modeling damage is one among many challenging aspects in computa-
tional mechanics which has led to a significant amount of research in the recent decades. One
can summarize the main questions into (1) when do cracks start to grow and (2) in which direc-
tion or where in the solid do they tend to propagate (see Fig. 1). In the early work of Griffith
[5], an energy criterion for crack propagation is mentioned which is well accepted for predicting
brittle fracture in materials/structures with an initial crack or defects. An alternative method
is proposed by Barenblatt [6] who introduced the the concept of cohesive fracture at the crack
tip. Here, in addition to fracture energy, the maximum strength of the material is treated as a
material property and used for predicting the crack nucleation.
Being able to differentiate between several phases through a smooth transition, the phase-
field (PF) damage model has shown a great potential to address damage in solids. This is
achieved by introducing an order parameter to describe the transition from intact material to
the fully damaged one [7, 8, 9]. The phase-field damage model has proven to be an elegant
tool, especially when multi-phases of a material [10], or a multiphysics problem [11, 12, 13] are
considered. For an overview of the model and a survey of recent advances, see [14].
Despite the interesting features of the PF damage model, one needs to treat the internal
length scale parameter with care. The length scale parameter controls the width of fracture
process zone and its value is usually small with respect to the structure’s size [15]. Utilizing the
PF damage model, same amount of energy is dissipated upon crack progress, independent of the
internal length scale parameter (see also [16] for further studies). Nevertheless, it was also shown
that this parameter directly influences the overall response of the structure (e.g. measured force-
displacement). Therefore, in the standard PF damage model, the length scale parameter can not
be seen as a pure numerical parameter [17, 18, 19], Utilizing a simplified analytical solution, one
can show the relationship between the length scale parameter and the strength of the material
[20, 21]. In other words, by utilizing a standard PF damage formulation, we are restricted in
choosing the length-scale parameter. This could be problematic when it comes to simulations
on a small scale as the relatively wide damage zone may create some boundary effects. Note
that since the value of the length scale parameter is linked to the material properties, one is not
to allowed to choose smaller values. Next, we look at the cohesive nature of fracture.
In addition to the fracture energy value, information on how fracture energy reaches its peak
2
value is essential. Taking the latter point into account, one is able to improve the standard PF
or even continuum damage models [2, 22]. Interestingly enough, such dependency is investigated
already in the context of cohesive zone (CZ) modeling [23, 24]. The constitutive relation for the
CZ model is defined by employing a so-called traction separation (TS) relation. It was shown
that CZ models can be calibrated based on information from lower scales down to atomistic
level [25, 26].
The above arguments are summarized in Fig. 1. On the left-hand side, a single notched
specimen is shown. The questions that we would like to address are when the crack starts to grow
and where in the solid it propagates (in which direction). It is shown that the fracture energy
is in general direction-dependent and there are in general certain preferential directions for the
crack propagation [1]. The latter point mainly determines the crack propagating direction. It
is also known that when the length of the initial defect L0 is large enough (compared to the
specimen size), the dominant factor in crack propagation is the fracture energy value Gc (or
fracture toughness [27]). On the other hand, when the initial crack length vanishes, the strength
of the material (also known as ultimate tensile stress), is the parameter that controls the crack
initiation.
Figure 1: Left: crack direction is under the influence of the direction-dependent fracture energy with an arbitrary
complex shape. Right: the ultimate stress of the material controls when the crack initiates [28, 22]
.
Interestingly enough, it was shown that the PF damage model can capture such a nonlinear
transition [29, 30]. On the other hand, the question remains on how can we treat the internal
length scale in this situation to make the formulation independent of it. The latter point will
3
be examined in what follows. Finally, the idea of this work is to combine the above-mentioned
features in one formulation to address anisotropic cohesive fracture in solids.
1.1. Phase-field modeling of cohesive fracture
By integrating the cohesive response of fracture in the PF damage model, it was shown that
one can omit the direct influence of the internal length scale parameter on the overall results.
Instead, one should introduce the maximum strength as an additional model parameter.
Available works on cohesive phase-field fracture are divided into two mainstreams. The first
category can be seen as an extension of the classical CZ model where instead of a sharp or
interphase interface one deals with a diffuse damage zone. Note that similar to CZ models, the
interphase position in these works is known in advance. Verhoosel and de Borst [22] included
the idea behind cohesive fracture in PF damage models by introducing an extra auxiliary field
for the displacement jump. See also [31, 32], where the authors described the sharp interface
by employing a diffuse zone which is the idea behind phase-field theory.
In the second category, the focus is on modifying standard PF damage models in a way
that they can represent the cohesive nature of fracture. Note that one is still able to predict
an arbitrary crack path using such methods. Motivated by [33, 2], new forms of energetic
formulations and functions were developed through which the cohesive fracture properties are
taken into account. These functions were recently used by [34, 35, 4] in PF damage models.
Wu and Nguyen [3] presented a length scale insensitive PF damage model for brittle fracture.
Utilizing a set of characteristic functions, the authors managed to incorporate both the failure
strength and the traction-separation relation, independent of the length scale parameter. Geelen
et al. [4] extended the PF damage formulation for cohesive fracture by making use of a non-
polynomial degradation function. The interested reader is referred to [36, 37] for more details.
1.2. Anisotropic fracture: direction-dependent fracture energy
Various microstructural features such as grain morphology or fiber orientation have a huge
impact on the material’s fracture properties. Such material features influence the crack direction
within an arbitrary solid. Anisotropic crack propagation can be linked to a direction-dependent
fracture energy function [38, 1]. Note that anisotropic elasticity is not enough to fully capture
anisotropic damage behavior [38, 39].
4
Anisotropic crack propagation in the context of phase-field models is based on two main-
streams. The first approach focuses on introducing structural tensors into the formulation which
act on the gradient of the damage variable, forcing the crack to propagate in certain directions
[40, 41]. This approach in consideration with only one scalar damage variable cannot model
arbitrary anisotropic fracture. Utilizing a second-order structural tensor will result in a fracture
energy distribution which only has one major preferential direction for the crack. Therefore, this
method is also known to cover only weak anisotropy. By utilizing higher-order damage gradient
terms and a fourth-order structural tensor, one can simulate the so-called strong anisotropy with
two preferential crack directions. The latter point is important in crystals with cubic symmetry
[42, 43, 40]. Beside being computationally more demanding, the latter approaches are limited
to certain shapes (distribution) of the fracture energy function. To take into account more
complicated fracture energy patterns, a promising extension would be to make use of several
damage variables. Nguyen et al. [44] introduced multiple PF damage variables. Each damage
variable is responsible for stiffness degradation in a certain direction (see also [45]). Neverthe-
less, this approach also increases the computational cost and opens up other questions, e.g. on
how different damage variables should influence the initial material’s elastic stiffness.
Interestingly enough, the (fracture) surface energy of a crystalline solid might become a
non-trivial function of orientation [46, 47, 48, 1]. Hossain et al. [49] presented the influence
of crystallographic orientation on toughness and strength in graphene. The latter observations
suggest that in general, one has to deal with an arbitrary complex distribution for the fracture
toughness of the material. Therefore, in the second strategy, the fracture energy parameter may
be defined as a function of the crack direction [1]. Very recently, the idea behind cohesive fracture
is also combined with anisotropic crack propagation using the PF damage model [50, 51, 52].
Still, fundamental studies are required on why the length scale insensitive PF damage model
might be necessary.
The outline of the current contribution is as follows. In section 2, the formulation for the
anisotropic insensitive phase-field damage model is discussed. In section 3, the discretization of
the problem for implementation in the finite element method is covered. Numerical examples
are then presented in section 4. Finally, conclusions and an outlook are provided.
5
2. Anisotropic phase-field model for cohesive fracture
In the left part of Fig. 2, the configuration of an anisotropic elastic body Ω is shown.The
specific material direction φ (e.g. fibers’ direction or grains’ orientation) is represented by the
vector a.
Figure 2: Configuration of a general elastic body and different applied boundary conditions.
According to the right hand side of Fig. 2, the direction-dependent fracture energy, strength
and elasticity of the material can be traced back to its material microstructure. The main
idea behind the current formulation is to take such properties into account in the PF damage
formulation. The position and displacement vector of an arbitrary point are represented by x
and u, respectively. The applied displacement, traction and body force vectors are denoted by
uex, tex and b, respectively.
The sharp crack Γc is represented by a diffuse damage field d(x). The width of the damage
6
zone is controlled by the length scale parameter lc. The internal energy density of the system ψ
is divided into an elastic part ψe and a damage part ψc. The latter shows the additional energy
of the newly created surfaces upon cracking:
ψ(ε, d,∇d) = ψe(ε, d) + ψc(d,∇d), (1)
where ε = 0.5 (∇u+∇uT ) is the strain tensor for small deformations. From Eq. 1, it becomes
clear that in the PF damage formulation, one deals with two separate fields, namely, the dis-
placement vector u and the damage parameter d. These two independent variables are strongly
coupled together. The elastic energy part takes the standard form
ψe =1
2ε : C : ε. (2)
The fourth order elastic stiffness tensor C is influenced by damage according to the following
split (see [17]):
C = fD C0 + (1− fD) P, (3)
C0 = λI ⊗ I + 2µIs, (4)
P = k0 sgn−(tr(ε)) I ⊗ I. (5)
Here, (Is)ijkl =1
2(δikδjl + δilδjk) is the symmetric fourth-order identity tensor. The second
order identity tensor is defined as (I)ij = δij. Considering Young’s Modulus E and the Poisson
ratio ν for elastic isotropic materials, λ =Eν
(1 + ν)(1− 2ν)and µ = G =
E
2(1 + ν)are the
Lame constants. In Eq. 3, the introduced damage function fD degrades the initial (undamaged)
material stiffness C0. The degradation function fD plays a significant role in the cohesive
behavior. According to [3, 2], for bilinear cohesive laws this function takes the form:
fD =(1− d)2
(1− d)2 + a1d + a1 a2 d2. (6)
In the above equation, a1 and a2 are constant model parameters that have to be chosen. They
are determined considering the cohesive properties of the model, e.g. the ultimate stress before
damage initiation and the value for strain at the fully broken state (see Appendix A and [3]).
Note that there are certainly other choices for the damage function as well (see for example [4]).
In general, the damage function takes the value one and when damage approaches one (crack is
fully developed), this function vanishes. The degradation function in Eq. 6 is plotted in Fig. 3
and compared against some classical choices.
7
Figure 3: Influence of parameters a1 and a2 on the degradation functions (Eq. 6). In the left and right hand
side we have a2 = −0.9 and a1 = 40, respectively
Furthermore, we require a split in tensile and compressive elastic energy parts to avoid
material damage under compressive loading. Here, the approach of [17] is considered, where
we take into account only the positive volumetric part of the strain to damage the material.
The sign function sgn−(•) = (• − | • |)/2, only takes the negative part of its argument. The
fourth-order projection tensor P is defined in Eq. 5 to exclude material parts in compression.
Moreover, the bulk modulus of the material is defined as k0 = λ +2
3µ. This approach is
simple to implement and yet effective in many applications especially when it comes to initially
anisotropic materials. There are more advanced splits available in the literature.
The energy for creating a new pair of surfaces per unit length is described as fracture energy
Gc. Therefore, in the PF damage formulation, we have [53, 17]∫Ω
ψc(d,∇d) dV =
∫Γc
Gc dA (7)
In this work, we take this concept further and make this energy dependent on the direction of
the crack, i.e. Gc(θ) = gc(∇d) [1]. Here it is assumed that the crack direction is perpendicular
to the damage gradient ∇d (see Fig. 2). Note that in the PF damage model, the damage
gradient vector can not easily be defined when there has no damage evolved in the system yet.
As will be discussed later, for a better convergence in the finite element calculations, we will
apply some numerical treatments. According to [1], the angle θ which represents the crack
direction is defined according to
θ = atan
(∇d · e2∇d · e1
)− π
2. (8)
8
In what follows, we also review the formulation for anisotropic PF damage model using
structural tensors. Note that for the latter approach one needs a constant value for the fracture
energy (i.e. Gc,0). The energy required for creating a crack Γc is regularized over the volume
such that we write [7, 53]:
ψc(d,∇d) =
ψc,s = Gc,0 γs(d,∇d), for structural anisotropy
ψc,a = Gc(θ) γa(d,∇d), for arbitrary anisotropy
, (9)
where γs and γa are the crack density function for the case of structural anisotropy and arbitrary
anisotropy, respectively. A more detailed definition is given in the following part. In Eq. 9, the
direction-dependent fracture energy is represented by Gc(θ).
Remark 1. The statement that the vector ∇d is orthogonal to the crack plane is an
approximation and does not hold in small regions at the crack tip. However, this effect is quite
localized. Although the influence of the latter point might be negligible but further studies on
this point would be interesting.
2.1. Modeling anisotropic fracture with structural tensors
To model anisotropic fracture, it is common in the literature to use a second-order structural
tensor. The crack density function in this particular case is written as:
γs(d,∇d) =1
c0 lcω(d) +
lcc0∇d ·A · ∇d. (10)
In Eq. 10, the scalar parameter lc is the internal length scale and represents the width of the
localized (damage) zone. Furthermore, the crack topology function is represented by ω(d). The
scalar parameter c0 = 4∫ 1
0
√ω(d)ds is obtained so the integration of the crack energy over
volume represent the material fracture energy Gc (see Eq. 7 and Appendix B). Similar to the
damage function fD, There are several choices for the crack topology function. For the cohesive
PF damage model, we focus on the following form, through which we will have c0 = π [3].
ω(d) = 2d− d2. (11)
The second order structural tensor A = I + α a ⊗ a which is constructed based on the
vector a, penalizes the crack direction at a certain angle [40]. This angle is in accordance with
the direction of the vector a = [cos(φ) sin(φ)]T . Therefore, one can write
A = I + α
cos(φ)
sin(φ)
[cos(φ) sin(φ)]
= I + α
cos2(φ) cos(α) sin(φ)
cos(φ) sin(φ) sin2(φ)
. (12)
9
In the above equation, the scalar parameter α determines the contribution of the preferential
directions in the energy term. In other words, the higher the parameter α is, the more energy
we require to form a crack perpendicular to the direction pointed by vector a. Moreover, the
angle φ denotes the preferred direction (e.g. grains, fibers and etc.). Since in this work we
focus on geometrically linear setting, the angle φ is kept constant through out the derivation
and further calculations.
After some simplifications (see Appendix C), one can obtain the following relation for the
anisotropic fracture surface energy utilizing the second-order structural tensor introduced in
Eq. 12:
ψc,s = Gc,0 γs =Gc,0
πlcω(d) +
Gc,0 lcπ||∇d||2
(1 + α sin2(θ − φ)
). (13)
Note that the angle θ = atan
(∇d · e2∇d · e1
)− π
2, is the crack direction and given in dependence of
∇d (see Eq. 8). The second term of crack surface energy in Eq. 13 is the response term for the
directional dependent fracture energy. Parameters Gc,0, α and φ are model input parameters.
This formulation is also known as a weak anisotropy [40]. Although in this work we will focus on
this particular formulation, later on, we will introduce the formulation with arbitrary anisotropy
as well.
Remark 2. Enhancing the crack density function with second-order structural tensors
showed a great performance in simulating anisotropic crack propagation in various applications.
However, by considering only one damage variable, such an extension is not general enough for
materials with strong anisotropy. Utilizing higher-order terms in the crack density function such
as γ =1
2lcd2 +
lc4∇d · ∇d+
l3c32∇2d : A : ∇2d is an interesting option. In the latter formula, A is
a fourth-order structural tensor which is defined employing preferable directions for the crack
(see [40]). These extensions can be even combined with several damage variables to take into
account more complex anisotropic behavior [54]. In what follows, we keep the crack density
function γ to be the same as a standard one and the amount of fracture energy is directly
plugged in through the function gc. This is motivated based on the arbitrariness of the fracture
energy for a solid (see [1]).
For the further derivation of the model, the following thermodynamic forces are introduced.
First, according to Eq. 2 and Eq. 3, the stress tensor as a conjugate force to the strain tensor
reads:∂ψe∂ε
= σ = C : ε = fD C0 : ε+ (1− fD) P : ε. (14)
10
Furthermore, the damage driving force Y from elastic energy reads:
∂ψe∂d
= −Y =dfDdd
1
2ε : Ch : ε, (15)
where Ch = C0 − P. By using the Euler-Lagrange procedure, the variational derivative of the
total energy with respect to the displacement field results in the standard mechanical equilibrium
[9, 40, 10]:
δuψ = ∂uψ − div(∂∇uψ) = 0⇒
div(σ) + b = 0 in Ω
σ · n = tex on Γt
u = uex on Γu
. (16)
Next, the variational derivative with respect to the damage field is considered [40, 10].
δdψ = ∂dψ − div(∂∇dψ) = 0⇒
Gc,0
lcπω′ − div
(lcGc,0
c0A : ∇d
)− Ym,s = 0 in Ω
∇d · n = 0 on Γc
(17)
In above equations, we utilize the maximum damage driving force Ym,s to consider for damage
irreversibility upon unloading. The expression for Ym,s is defined as maximum value between the
undamaged elastic strain through the simulation time (ψ0e(t)) and the damage energy threshold
(ψth) [3, 52]:
−Ym,s = f ′DHs = f ′D maxt
(ψ0e(t), ψth,s). (18)
Here, the maximum value of stored undamaged elastic energy ψ0e =
1
2ε : Ch : ε during the
simulation time is denoted by Hs. The scalar parameter Hs = maxt(ψ0e(t), ψth,s) is treated as a
history variable throughout the simulation (see also [9, 55, 56, 3]).
The energy threshold ψth,s ensures that damage remains zero as long as the elastic energy
of the system is below this threshold. This is achieved based on the linear damage term in the
damage topology function ω(d) (Eq. 11). See also section 2.3 and [56, 3]. The system’s elastic
energy right before onset of failure can be written in terms of the failure initiation strain ε0 or
the failure stress σ0:
ψth,s =1
2E ε20 =
1
2Eσ20. (19)
More explanations for the chosen relations in the above equation are provided at the end
of this section. For the derivative of the damage function fD with respect to to the damage
11
variable, we have
f ′D =dfDdd
=−2(1− d) (a1d + a1 a2 d
2)− (1− d)2 (a1 + 2a1 a2 d)
((1− d)2 + a1d+ a1a2d2)2 . (20)
See also Eq. 6 and explanations provided afterwards for parameters a1 and a2. Based on the
studies of [35, 3], to represent a softening behavior similar to the bi-linear cohesive zone model,
we choose the following form for these constant:
a1,s =4E Gc
πlc σ20
, a2,s = −0.5. (21)
Remark 3. The scalar parameter a1 in Eqs. 6, 21 and 29 is defined to be length-scale
dependent. As we will show later, this is one main reason why we have length-scale insensitive
results for our cohesive phase-field damage model. In other words, via such a formulation one
can control the maximum strength of the new material property (input) σu = σ0.
2.2. An arbitrary anisotropic fracture energy
For this formulation, the crack density function γa, takes the standard form
γa(d,∇d) =1
c0 lcω(d) +
lcc0∇d · ∇d. (22)
Similar descriptions as for the previous case hold here for the parameter c0 = π, the length
scale parameter lc as well as the crack topology function ω(d). Based on the recent work of
the authors presented in [1], to model anisotropic crack propagation, one can directly apply
an arbitrary shape for the fracture energy function. Considering the crack angle θ (Eq. 8), it
is suggested that the direction-dependent fracture energy function Gc(θ) can be obtained by
summation over the frequency energy function. Here the sub-index m which belongs to natural
numbers represents the frequency number:
Gc(θ) =∑m
κm(1 + αm sin2 (m(θ + θ′m))
), m ∈ N. (23)
The angle θ represents the crack direction and the latter is perpendicular to the vector ∇d (see
Eq. 8). Parameters κm, αm and θ′m are model input parameters. To be able to compare it to
the case of weak anisotropy using a second-order structural tensor, we will particularly consider
only one energy frequency (m = 1, κ1 = Gc,0, α1 = α and θ′1 = −φ). The simplified version of
the crack-free energy is written as:
ψc,a = Gc(θ) γa = Gc,0
(1 + α sin2(θ − φ)
)( 1
πlcω(d) +
lcπ||∇d||2
). (24)
12
Interestingly enough, there are similarities between the current methodology and the modifi-
cation for the anisotropic crack density function introduced by [57]. Eq. 24 shares a lot of
similarities with the expression in Eq. 13, although they are not exactly the same.
Since the elastic part of the energy remains as before, the definition for the stress tensor and
damage driving force is the same as described in Eq. 14 and Eq. 15, respectively. Therefore, using
the Euler-Lagrange procedure, the variational derivative of the total energy with respect to the
displacement field results in the same expression described in Eq. 16. Based on the crack density
function with arbitrary direction-dependent fracture energy, and considering Gc(θ) = gc(∇d),
for the variational derivative with respect to the damage field we have [1]
δdψ = 0⇒
gc(∇d)
lcπω′ − div
(lc gc(∇d)
c0∇d)− div(γgd)− div(sdHa)− Ym,a = 0 in Ω
∇d · n = 0 on Γc
(25)
Similar to Eq. 18, the expression for Ym,a is defined as maximum value between the undamaged
elastic strain through the simulation time (ψ0e(t)) and the new damage energy threshold (ψth,a):
−Ym,a = f ′DHa = f ′D maxt
(ψ0e(t), ψth,a). (26)
We choose the following definition for damage threshold (see also [56]):
ψth,a =1
2Eσ2u(θ). (27)
For the direction-dependent tensile strength σu(θ), we propose the following function:
σu(θ) =∑m
σ0,m(1 + αm sin2 (m(θ + θ′m))
)pm, m ∈ N. (28)
Similar to the direction-dependent fracture energy, here m denotes the frequency number. The
total strength of the material is the summation over all the active frequencies. Furthermore, pm
denotes an additional material parameter in this work. The structure of Eq. 28 is also motivated
by the work of [49] and certainly can be modified according to specific application. Utilizing
Eq. 28 allows us to have a directional maximum tensile strength. It worth mentioning that the
other parameters such as αm and θ′m, are the same as the ones in Eq. 23.
For the case of arbitrary direction-dependent fracture energy, the following relations are
proposed to obtain the constants a1 and a2 in the damage function fD (see Eqs. 6 and 20).
a1,a =4E Gc(θ)
πlc σ2u(θ)
, a2,a = −0.5. (29)
13
The two new terms in Eq. 25, gd and sd, are imposed by the directional dependency of the
fracture energy function and the degradation function, respectively (compare Eq. 25 to Eq. 17
and see [1]). Finally, we have the following definitions for the new terms in Eq. 25:
gd =∂Gc(θ)
∂∇d=∂Gc(θ)
∂θ
∂θ
∂∇d, (30)
sd =∂fD∂∇d
=∂fD∂a1
∂a1∂θ
∂θ
∂∇d. (31)
For the calculation of new terms gd and sd in Eq. 25, the following steps have to be taken:
∂Gc(θ)
∂θ= Gc,0 αm sin(2m(θ + θ′)), (32)
∂fD∂a1
=(1− d)2(d− d2/2)
[(1− d)2 + a1d+ a1a2d2]2 , (33)
∂a1∂θ
=4E Gc
π lc σ20
αm(1− 2 pm) sin(2m (θ + θ′))(1 + α sin2(θ + θ′)
)2pm , (34)
∂θ
∂∇d=
1
||∇d||2
−∇d · e2∇d · e1
. (35)
Remark 4. The new terms mentioned in Eq. 30 and Eq. 31 are the contributions by
considering an arbitrary shape for the direction-dependent fracture energy function as strength.
As we will discuss in the next section, these terms can be computed explicitly within the finite-
element calculation to reduce the complexity of the implementation (see also [1] and Algorithm
1).
2.3. Explanation of damage threshold
Having a linear term in the crack topology function ω(d) enables us to have an initial elastic
stage before damage initiation. In other words, by considering the threshold, one can guarantee
that the value of damage remains zero (d = 0) in Eq. 17. A simple one-dimensional analysis
is carried out for clarification. Considering a uniform distribution for the damage variable
(d′ = ∂d/∂x = 0), the governing equation for damage (Eq. 17) reduces to
Gc
lcπ(2− 2d)− f ′DH = 0. (36)
14
Note that if there is no damage threshold ψth, damage takes the value one. Considering Eqs. 18
and 27, one can further simplify the damage governing equation to
Gc
lcπ(2− 2d)− a1
1
2Eσ20 = 0. (37)
Having Eq. 21 in hand, the above expression guarantees that the damage value remains 0 before
the threshold is met. After passing the threshold (i.e. ψ0e > ψth), the history parameter H in
Eq. 36 is replaced by ψ0e =
1
2ε : Ch : ε which derives the damage to evolve.
2.4. Cohesive zone model
Here we summarize the formulation of the cohesive zone model (CZM). The CZM relates
the traction vector t = [tn, ts]T to the displacement jump or gap vector g = [gn, gs]
T :
tn = k0 (1−D) gn, (38)
ts = β2k0 (1−D) gs. (39)
Here, k0 is the initial stiffness of the cohesive zone model. Damage at the interface (D) is
determined based on the introduced traction-separation relation [24, 58]:
D =
0 if λ < λ0
λfλf − λ0
λ− λ0λ
if λ0 < λ < λf
1 if λf < λ
. (40)
The parameter λ =√〈gn〉2 + (βgs)2 represents the amount of separation at the interface with
gn and gs being the normal and shear gap vector, respectively. The parameters of the model are
summarized as (1) the maximum strength of the interface t0 = k0λ0, (2) the critical separation
where damage starts λ0, (3) the final separation at which the traction goes to zero λf , and
(4) the parameter β which governs the contribution of the separation in shear direction. As a
result, the interface fracture energy is computed using Gc,int =1
2t0λf .
2.5. Summary of different formulations
Here we would like to compare different formulations for the convenience of the reader.
First off, we have the comparison between modeling anisotropic fracture utilizing structural
tensor and arbitrary direction-dependent fracture energy in Table 1. Note that both of these
anisotropic formulations are based on cohesive fracture models [2, 3, 4].
15
For the sake of completeness, a comparison is also performed between the standard PF
damage model and cohesive PF damage model in Table 2. The standard phase-field model
which is used in this work is based on the so-called AT-2 (see also [9]). Readers are also
encouraged to see Appendix A and B. The formulations in Table 2 can be simply coupled with
those in Table 1 to construct anisotropic cohesive phase-field models.
Structural tensors Arbitrary fracture energy function
Crack energy ψc,s = Gc,0 γs(d,∇d) ψc,a = Gc(θ) γa(d,∇d)
γs =ω(d)
c0lc+lcc0∇d ·A · ∇d γa =
ω(d)
c0lc+lcc0∇d · ∇d
Gc,0 = const. Gc(θ) =∑
m κm(1 + αm sin2 (m(θ + θ′m))
)Damage function fD =
(1− d)2
(1− d)2 + a1d+ a1a2d2fD =
(1− d)2
(1− d)2 + a1d+ a1a2d2
a1,s = (4EGc)/(πlc σ20) a1,a = (4EGc(θ))/(πlc σ
2u(θ))
a2,s = −0.5 a2,s = −0.5
σ0 = const. σu(θ) =∑
m σ0,m(1 + αm sin2 (m(θ + θ′m))
)pmDamage threshold ψth,s =
1
2Eσ20 ψth,a =
1
2Eσ2u(θ)
Table 1: Comparison between anisotropic damage models based on structural tensors and arbitrary fracture
energy function.
16
Standard phase-field model Cohesive phase-field model
Crack topology function ω(d) = d2 ω(d) = 2d− d2
c0 = 2 c0 = π
Damage function fD = (1− d)2 fD =(1− d)2
(1− d)2 + a1d+ a1a2d2
— a1,s = (4EGc)/(πlc σ20), a2,s = −0.5
History parameter H = maxt(ψ0e(t)) H = maxt(ψ
0e(t), ψth)
— ψth =1
2Eσ20
Table 2: Standard versus cohesive phase-field damage models.
3. Weak form and discretization
Through the FE discretization procedure, the following approximation for displacement and
damage fields within a typical element and their derivatives are employed (see [59, 60]).u = Nuue
d = Ndde
,
ε = Buue
∇d = Bdde
. (41)
The subscript e represents the nodal values of the corresponding quantity. Utilizing linear
shape functions and considering a quadrilateral 2D element, one obtains the following matrices
for shape functions and their derivatives in N and B matrices, respectively:
Nu =
N1 0 · · · N4 0
0 N1 · · · 0 N4
2×8
, Nd =[N1 N2 N3 N4
]1×4
, (42)
Bu =
N1,x 0 · · · N4,x 0
0 N1,y · · · 0 N4,y
N1,y N1,x · · · N4,x N4,y
3×8
, Bd =
N1,x N2,x N3,x N4,x
N1,y N2,y N3,y N4,y
2×4
. (43)
Next, the weak form of the governing Eqs 16, 17 and 25 are obtained. After applying the
introduced discretization form, the residual vectors for the Newton-Raphson solver are obtained
17
for the displacement and damage field:
ru = −[(∫
Ωe
BTuCBuue −NT
u b
)dV −
∫∂Γt
NTu t dA
]8×1
(44)
rd,s = −[∫
Ωe
2Gc,0lcπ
BTdABdde +NT
d
(ω′(d)Gc,0
c0lc− Ym,s
)dV
]4×1
(45)
rd,a = −[∫
Ωe
2gc(∇d)lcπ
BTdBdde +NT
d
(ω′(d)gc(∇d)
c0lc− Ym,a + γBT
d gd +H iaB
Td sd
)dV
]4×1(46)
The above equation are the element residuals. The residual vector for the case of structural
and arbitrary anisotropy are denoted by rd,s and rd,a, respectively. Note that either rd,s or rd,a
is used in what follows. As shown in Algorithm 1, we utilize a staggered approach which is
known to be able to handle the instabilities upon damage progression in a more robust way
[61, 62]. The superscript emphasizes the semi-implicit scheme which is used for computing the
direction of the crack. The residuals and stiffness matrix at the element level are shown in
Algorithm 1. Here, the solver finds the solution at time i + 1 using an iterative approach till
∆ui+1e,k+1 = ∆Di+1
e,k+1 = 0. The letter k represents the Newton iteration number.∆ui+1k+1
∆di+1k+1
=
ui+1k+1 − u
i+1k
di+1k+1 − d
i+1k
= −
Ki+1uu,k+1 0
0 Ki+1dd,k+1
−1 Ri+1u,k+1
Ri+1d,k+1
. (47)
In the above equation, Ri+1 and Ki+1 denote the assembled global residual vector and stiffness
matrices, respectively.
18
Inputs: die, uie, ||∇dc|| and material properties
Outputs: di+1e and ui+1
e
∇di = Bddie → θi = tan−1
(∇diy∇dix
)if ||∇di|| ≥ ||∇d||c then
Compute gic (Eq. 23) and σiu (Eq. 28)
Compute gid (Eq. 30) and sid and (Eq. 31)
else
Set gic = Gc,min and σiu = σu,min
Set gid = 0 and sid = 0
end
Compute ψi+1e =
1
2εi+1 : Ci : εi+1 (Eq. 2) and ψith,a (Eq. 27)
Compute H ia = max(ψie, ψ
i+1e , ψith,a) (Eq. 26) and ai1,a (Eq. 29)
ri+1u =
∫Ωe
(BTuC
iBuui+1e −NT
u b)dV +
∫Γt
NTu t dA
ri+1d =
∫Ωe
(2giclcπBTdBdd
i+1e +NT
d
(ω′(di+1
e )gicπlc
+ f ′D(di+1e )H i
a
)+ γBT
d gid +H i
aBTd s
id
)dV
ki+1uu =
∫Ωe
BTuC
iBu dV
ki+1dd =
∫Ωe
(2lcg
ic
πBTdBd +NT
d
(ω′′(di+1
e )gicπlc
+ f ′′D(di+1e )H i
a
)N
)dV
Algorithm 1: Element residual vector and stiffness matrix for the case of arbitrary anisotropy
Algorithm 1 is written at the element level. As discussed in [1], the parameter ||∇d||cis introduced since at the beginning of the simulation there is no damage to determine the
direction-dependent property based on ∇d. According to studies in [1], its value should be
large enough for avoiding convergence issues. We will provide suggestion for choosing this
parameter it what follows. Note that the explicit evaluation of gd and sd, causes the vanishing
of these terms in element stiffness.
4. Numerical examples
The material parameters used for the following numerical studies are reported in Table 3.
We will focus on damage propagation in an elastic solid with an initial crack. Note that for the
first set of studies, the elastic constants are not rotated according to the preferential fracture
19
direction (i.e. we have initially isotropic material). The anisotropic elastic properties will
influence the crack direction as well (see [63] for such studies). By focusing on initially isotropic
material, one can better focus on the influence of the direction-dependent fracture energy on
the crack path. Further studies on the combined influence of anisotropic elasticity and fracture
are postponed to future studies.
Unit Value
Lame’s Constants (λ, µ) [GPa] (132.6, 163.4)
Fracture energy Gc = Gc,0 [J
m2] ≡ [GPa.µm]103 40
Ultimate strength σ0 = σ0,1 [GPa] 5
Damage internal length lc [µm] 0.025− 0.2
Frequency number m [-] 1, 2
Fracture energy parameter αm [-] 0.0, 3.0
Fracture energy parameter θ′m [-] −40, 0
Structural parameter α [-] 0.0, 12.0
Structural parameter φ [-] −40, 0
Damage parameter ||∇d||c [-] 0.2
Material strength parameter pm [-] 0.1
Table 3: Parameters for the anisotropic PF damage formulation.
4.1. Crack propagation in an initially isotropic solid
According to Fig. 4, a single notched specimen is studied. Simulations are carried out in a
2D configuration based on a plane-strain assumption. Two different dimensions are chosen for
numerical studies (see Table 4). Geometry B is constructed by scaling geometry A by the factor
1/8. As will be shown, choosing the smaller geometry will help us to motivate and understand
better the idea behind cohesive fracture. Moreover, on the right-hand side of Fig. 4, the mesh
topology is illustrated. In all simulations, we make sure that enough elements are utilized
depending on the chosen value for the length scale parameter lc.
20
geometry A geometry B
Length in x direction Lx, [µm] 4.0 0.5
Length in y direction Ly, [µm] 4.0 0.5
Initial crack length L0, [µm] 2.0 0.25
Table 4: Chosen dimensions for the numerical studies. The geometry A is 8 times larger than the geometry B.
Figure 4: Boundary conditions and geometry of a single notched specimen.
We will start by assuming a constant fracture energy value also known as isotropic crack
propagation. For the defined boundary value problem in Fig. 4, the crack propagates along
a horizontal line without any deviation. The system with geometry A is simulated utilizing
different models.
In Fig. 5, the results of the standard phase-field (SPF) damage, cohesive phase-field (CPF)
damage as well as cohesive zone (CZ) model are presented in different rows (see also Table 5).
In the simulation using the SPF damage model, the internal length scale parameter is set to
lc = 0.05 µm. The latter value is chosen based on the available analytical relations between the
internal length scale and other material properties (see [28]):,
σ0 =9
16
√E Gc
3 lc⇒ lc ≈ 0.05 µm. (48)
21
Abbreviation Model
SPF Standard Phase-Field
CPF Cohesive Phase-Field
CZM Cohesive Zone Model
Table 5: Summary of the models utilized in this work.
In the last row of Fig. 5, the same boundary value problem is calculated utilizing the standard
bi-linear CZ model [24]. Since we know that the crack propagates in the horizontal direction,
CZ elements are introduced accordingly.
For the first study, the interface behavior is assumed to be isotropic, i.e. β = 1. The CZ
parameters such as the maximum strength of the CZ model (t0), the undamaged stiffness of
cohesive zone model (k0), and the area beneath the TS curve (Gc,int) are chosen to represent
very similar material properties reported in Table 4. Therefore, λf = 0.016 µm is obtained.
Moreover, the CZ initial stiffness is set to k0 = 5× 1012 [GPa
µm] to get the closet possible result
to the phase-field approach.
Comparing the results obtained from SPF and CPF for the larger geometry does not show
any obvious difference. In other words, when the dimension of the problem (Lx) is comparatively
larger than the internal length scale (lc), the SFP performs well enough. The latter point is well
accepted in many engineering applications and, therefore, motivated many researchers to treat
the parameter lc as a material parameter. On the other hand, when it comes to geometry B,
utilizing SPF results in a wide spread of the damage zone. Although the same internal length
scale parameter is used for the simulation with CPF, the damage zone is much more localized
in a certain region (see the idea of the threshold for damage introduced in Section 2.4).
22
Figure 5: comparing the crack paths of CPF, SPF and CZM models (for the isotropic damage case)
Remark 5. The spread of the damage zone for the case of the SPF formulation is only
problematic, if the geometry is relatively small. One remedy is decreasing the internal length
scale which leads to a narrower zone. However, by doing so, we will change the basic material
properties that we have (i.e. maximum tensile strength) which is not allowed. Utilizing the
23
CPF formulation, one can select smaller values for lc depend on the dimension of the problem.
The total reaction force obtained from the calculations versus the applied displacement at
the top edge is plotted in Fig. 6. For geometry A, with larger sizes (top row), one observes
the typical sharp drop upon sudden and brittle fracture. The results obtained by using the CZ
model matches also very well with the SPF models.
Figure 6: Comparing the response of CPF, SPF and CZM models for the isotropic damage case. The upper row
is related to geometry A with larger sizes and the lower one is related to geometry B.
Using SPF and decreasing the parameter lc, the peak point of the reaction force increases
as expected. In other words, one can fit the lc parameter such that the peak point matches well
with the results of the CZ model. Using the CPF model, the values for the reaction force are
almost insensitive with respect to the internal length scale lc (see also [3] for similar results).
This is due to the fact that more information about the fracture property is now included
in the model (namely the strength of the material t0 which is not the case for SPF models).
24
Interestingly enough, the results of the CPF model are pretty much following the CZ model
which confirms our latter statement.
Focusing on the results of geometry B, one notices a smooth transition in the reaction force
after the maximum load peak is reached. The latter observation only holds for the CPF and the
CZ models. Note that, we store less elastic energy in geometry B with a smaller size compare
to geometry A. In other words, it will be easier for the system to dissipate this total energy by
means of crack propagation. Similar to the previous case, the results of the SPF formulation
show a clear sensitivity with respect to the length scale parameter lc, while the CPF formulation
is not only almost insensitive but also matches very well with the CZ model results.
Remark 6. Due to size difference, the stored elastic energy is much higher in geometry A
than in geometry B. Since the crack surface cannot dissipate all this energy, we observe a sudden
drop in the reaction force plots. The sudden drop is due to the staggered algorithm which is
used in this study. One may use other techniques like the arc-length method to capture snap-
back for geometry A [64, 59, 27]. Similar behavior is expected using artificial viscous parameter
in solving the system of equations [9, 24].
4.2. Anisotropic crack propagation utilizing structural tensor
We look at anisotropic cracking in specimens described in Table 4, now utilizing the formu-
lation based on the structural tensor A (see Eqs. 13, 17 and 45). The model parameters are
reported in Table 3. Note that by utilizing a structural tensor one can obtain the equivalent
fracture energy distribution as a function of the crack angle. Here, by setting α = 12 the ratio
between the maximum and the minimum energy value is equal to 3.0. This ratio will be used
directly in further studies.
25
Figure 7: Studies on anisotropic crack propagation using standard and cohesive PF damage formulation utilizing
different length scale parameters lc. Here, the geometry A is used.
Figure 8: Comparing the response of CPF, SPF and CZM models for the anisotropic damage case.
26
The fracture energy distribution in the polar coordinate is also plotted in the correspond-
ing figures (see white peanut-shaped curves in Fig. 7). According to Fig. 7, the influence of
the length scale parameter lc on the obtained crack path is studied. It seems that for both
approaches, the crack path angle converges to a certain value θc ≈ 30. By increasing the
parameter α, the angle θc converges to the preferential crack direction φ [1, 52, 40]. Similar to
the isotropic case, the crack path obtained by SPF and CPF are very close together. Moreover,
for a given length scale lc, the damage zone using SPF is relatively wider compared to CPF.
The reaction forces for the aforementioned simulations are shown in Fig. 8. For the case of
SPF, the reaction forces indicate the dependency of the strength to the length scale parameter.
On the other hand, utilizing the CPF, the obtained reaction forces are very much similar with
just a slight increase in the peak force. For anisotropic media, one can still use the simplified
analytical solutions to relate the maximum strength of the material to the internal length lc.
We will try to address this point in the next part.
Next, we focus on geometry B, where the specimen dimensions are relatively small and closer
to the chosen length scale parameter lc. The crack paths using SPF and CPF are pretty much
similar, even for the case of anisotropic fracture energy. Therefore, in Fig. 9, only the results of
the CPF model are shown. Due to the new geometry dimensions, the final crack path slightly
changes to θc ≈ 20. Note that the material properties such as preferential crack direction φ are
the same as before. Nevertheless, the amount of stored elastic energy and its competition with
the crack energy determines the final crack path which is different compared to geometry A.
In the next step, we studied the same anisotropic cracking utilizing the CZ model. Here, we
take advantage of the PF fracture results to determine in which direction the crack propagates
(θc ≈ 20). The very same plane is enriched with CZ elements introduced previously. Moreover,
we also studied the influence of fracture mode-mixity. In other words, the parameter β in the CZ
formulation (see Eq. 39) is varied. Choosing a relatively small value, i.e. β = 0.01, indicates a
weak contribution from the shear direction upon shear opening. Furthermore, choosing β = 1.0
means isotropic behavior for the CZ formulation. Finally, by setting β = 100, the contribution
of the shear traction is much more pronounced.
27
Figure 9: Top: studies on anisotropic crack propagation by using cohesive phase-field damage formulation
utilizing different length scale parameter lc. Here, geometry B is used and the material properties are the same
for all these studies. Bottom: Simulation results using CZ with different values for β.
The results of the comparison between the CPF and CZ model with different mode-mixity
parameters are summarized in Fig. 10. First, we observe that the results of the CPF model are
almost length-scale insensitive even in the case of anisotropic fracture. Second, the results of
the CZ model match very well with the CPF model only if β = 1.0 or β = 0.01. In other words,
when the contribution of the shear traction is much more due to the fracture mode-mixity (i.e.
β = 100), the post-fracture results of the CZ model deviate from those of CPF. The later point
opens up the necessity of taking into account the mode-mixity into the PF damage formulation.
28
By doing, one can perhaps can think about utilizing more damage variables for each fracture
mode [65]. See also [66, 67].
Remark 7. Despite being consistent with the CZ models, the cohesive phase-field formu-
lation still lacks one of the main features of CZ models which is the mode-dependent nature of
the fracture. The latter point should be studied in future developments. One idea would be to
consider multiple damage variables to represent different fracture modes [65].
Figure 10: Comparison between CZ with different β and CPF.
4.3. Anisotropic crack propagation utilizing a direction-dependent fracture energy
In this section, we look at anisotropic cracking in specimens described in Table 4, this time
by utilizing the formulation based on arbitrary anisotropy (i.e. Eqs. 24, 25 and 46). The model
parameters regarding anisotropic fracture (αm and θm) are reported in Table 3. Furthermore,
in the current simulations we propose ||∇d||c = 0.04 lc. It is checked that this parameter is
small enough so the numerical solver converges and the obtained results remain unchanged with
respect to this (see also studies in [1]).
The results of the obtained crack path are plotted in Fig. 11 for different values of the length
scale parameter. Similar to the previous study, the crack path angle converges to a certain value
θc ≈ 30 for all the cases. The reaction forces are shown in the lower part of Fig. 11. Interestingly
enough, utilizing the CPF model, the obtained reaction forces are very similar which shows the
almost insensitive response of the formulation with respect to the length scale parameter.
29
Figure 11: Studies on anisotropic crack propagation using the introduced anisotropic CPF damage formulation.
These results are obtained based on an arbitrary function of the fracture energy as well as
the strength of the material. In other words, with one single damage variable one can take into
account any complicated fracture energy distribution. The latter point is an efficient way to
simulate anisotropic cracking in many available materials (see also [1]).
Remark 8. Despite the nearly length scale insensitive results we cannot simply choose lc as
large as we want. According to [68, 3], for having numerical stability, a1 ≥3
2should be fulfilled,
which means lc ≤ 0.85 lch. Here, lch =EGc
σ20
is the characteristic length scale of the problem.
Therefore, there is an upper limit for lc.
30
4.4. Three-point bending test with anisotropic properties
The geometry and material parameters for this test are shown in Fig. 12. The material
direction which can be interpreted as fibers direction or a layered material is represented by
the angle φ = 30. For more realistic calculations, the anisotropic elastic properties are also
considered for this example by having the grain orientation depicted in Fig. 12. The anisotropic
elastic and fracture properties are reported in Table 6, see also [69]. This problem is solved
utilizing the introduced anisotropic CPF model with arbitrary function for the fracture energy
distribution (similar to Section 4.3).
Unit Value
Elastic constant C11 [MPa] 142350
Elastic constant C12 [MPa] 188782
Elastic constant C16 [MPa] 115880
Elastic constant C26 [MPa] 192680
Elastic constant C22 [MPa] 321110
Elastic constant C66 [MPa] 126370
Fracture energy Gc,0 [J
m2] ≡ [MPa.mm]103 54
Ultimate strength σ0 = σ0,1 [MPa] 10
Damage internal length lc [mm] 0.4− 1.0
Frequency number m [-] 1
Fracture energy parameter αm [-] 3.5
Fracture energy parameter θ′m [-] 60
Damage parameter ||∇d||c [-] 0.2
Material strength parameter pm [-] 0.1
Table 6: Parameters for the anisotropic PF damage formulation and anisotropic material.
A displacement on the top edge is applied and the reaction forces are measured accordingly.
As expected, due to the anisotropic properties the crack direction runs along the angle φ. For
the chosen material properties, the obtained crack path is very close to this preferential crack
angle, i.e. θc ' 30. We also study the influence of the length scale parameter lc. For the chosen
values, not only the final crack paths but also the overall measured reaction forces are in very
31
good agreement (see the lower part of Fig. 12). To ensure the accuracy of the obtained results,
a mesh convergence study is performed for the case with lc = 2 mm. See also similar studies in
the context of rock mechanics [70] and also when it comes to fiber composite materials [71, 72]
utilizing standard PF damage models.
Figure 12: Studies on the three-point-bending test. For two different length scale parameters lc, the results
regarding the obtained crack paths as well as overall reaction forces are compared.
It is worth mentioning that, depending on the chosen length scale parameter, the element
size for the finite element calculation can be changed to reduce computational time. In the
current studies the computational cost of the simulation with lc = 1.0 mm is almost half the
case with lc = 0.4 mm. The latter point shows another advantage of the length scale insensitive
formulation and its flexibility for choosing the length scale parameter. Nevertheless, one has to
32
keep in mind the restrictions described in Remark 8. Depending on the size of the specimen,
this parameter should not be chosen too large, otherwise the crack path might get too diffused
which may not be physically accurate.
4.5. Anisotropic cracking in crystalline materials with diffuse interphase
To show the potential of the cohesive phase-field approach, we discuss the cracking in a
simple bi-crystalline system according to Fig. 13. Here, the grain boundary is represented by
a diffuse zone in green color. The two neighboring grain each has specific orientation as shown
in the figure. For the diffuse interphase, an anisotropic distribution for the fracture energy is
considered which its orientation is exactly set according to the grain boundary angle (i.e. 75).
All the anisotropic cohesive phase field formulations are based on structural tensor according to
Eqs. 13, 17 and 45. Other material properties such as elastic modulus and fracture properties
are according to Table 7.
Unit Value
Lame’s Constants (λ, µ) [GPa] (132.6, 163.4)
Bulk fracture energy Gc,b [GPa.µm]103 30
Interphase fracture energy Gc,ip [GPa.µm]103 30, 15
Bulk ultimate strength σ0,b = σ0,1 [GPa] 4
Interphase ultimate strength σ0,ip = σ0,1 [GPa] 4, 2
Damage internal length lc [µm] 0.025
Structural parameter α [-] 12.0
Structural parameter φ [-] −30, 30
Table 7: Parameters for the anisotropic PF damage formulation.
Note that in this example, there is no need for the insertion of additional cohesive zone
elements. In other words, the cohesive phase-field approach on its owns includes the same
properties. By applying displacement in the vertical direction on the top edge, crack propagation
in this system is studied. For similar studies readers are encouraged to see [73, 74, 1, 75].
For a better comparison, the fracture energy value for the interphase Gc,ip is varied against
the one for the bulk part Gc,b. In the middle part of Fig. 13, the results for the weaker grain
boundary are shown, where the crack tends to propagate along the interphase and then goes to
33
the other grain. On the other hand, by increasing the interphase fracture energy, as shown in
the right-hand side of Fig. 13, the transgranular fracture is observed.
Figure 13: Studies on anisotropic cohesive fracture within a bi-crystalline system. The grain boundary is treated
as a diffuse zone. The introduced anisotropic cohesive phase filed model can handle cracking within the bulk
and interphase.
5. Conclusion and future work
In this contribution, we try to address anisotropic cohesive fracture using the phase-field
damage model. In other words, direction-dependent damage initiation and propagation within
an arbitrary anisotropic solid are under focus.
It is well established that standard PF damage models provide a consistent formulation
that can predict crack initiation and propagation. By dissipating the fracture energy within a
diffuse zone controlled by the length scale parameter, these models solve the problem of mesh
sensitivity during damage progression. The length scale parameter, on the other hand, has a
significant influence on the global response of the model. This parameter is shown to be related
to the maximum strength of the material and, therefore, can control damage nucleation. We
discuss that the latter point is not desirable for all applications, especially when the size of the
specimen is not large enough compared to the internal length scale parameter. Furthermore, for
high-strength materials, the mesh has to be extremely refined which increases the computational
costs significantly.
34
Firstly, the sensitivity of the system’s global response with respect to the length scale pa-
rameter is shown for standard PF damage formulation. Secondly, an insensitive formulation
[2, 3, 4] is adopted and then extended for the anisotropic case. In particular, we focus on
utilizing the direction-dependent fracture energy formulation [1] and second-order structural
tensors [40]. Considering the numerical implementation, a linearization procedure and details
of utilized algorithms are discussed as well. The crack initiation and propagation in a single
notched specimen with two different geometries as well as a simple three-point bending test are
studied. It is shown that the formulation can produce almost insensitive results with respect
to the length scale parameter for both isotropic and anisotropic cases. We also compared the
numerical results against results obtained by studying cracking using the standard cohesive-
zone model. It is shown that the framework can reproduce the results from the CZ formulation,
especially when there is no severe difference between different opening modes behavior.
We conclude that the cohesive phase-field formulation has two main advantages: we include
more (clear) physics into account by introducing the strength and fracture energy as input
parameters. In other words, the length scale parameter can be treated as a numerical parameter
which should be small enough, depending on the application and the boundary value problem.
The latter point is extremely helpful in multiphysics problems where the fracture properties are
under the influence of other fields [76]. Furthermore, one can relatively increase the mesh size.
We show that the latter point reduces the computational time significantly without any severe
change in the predicted crack path or overall obtained load-displacement curves.
The developed model can be applied in efficient numerical modeling of fracture at smaller
scales. For example, see studies by [77, 58] on micro-coating layers where the thickness of the
coating system is about a few micrometers which are in order of the obtained internal length.
Therefore, it not so easy to simulate the problem with standard PF damage models.
Further comparisons with similar models can be very interesting to complete our understand-
ing of the anisotropic nonlocal fracture in solids. As an example, the PF damage formulation
benefits from regular mesh generation, while cohesive zone models suffer from predefined crack
path zones and a specific mesh algorithm. Also, comparisons with other methodologies such as
XFEM and Peridynamics would certainly be interesting.
Apart from the advantages of the current anisotropic CPF formulation, there are some open
issues and possibilities for further improvements. We showed that CPF models still lack to
35
capture mode-dependent fracture properties. A crucial enhancement for the formulation could
be made to consider different modes of opening. As a possible remedy, one could introduce
different damage variables for each opening mode. Utilizing multiple damage variables would
cause a degrading of the elasticity matrix components with different damage values. Mean-
while, multiple damage variables could be beneficial and enable the model to capture different
stresses and anisotropic responses [78, 65]. Another idea for further developments would be
to degrade the fracture toughness value to represent fatigue behavior [79]. Finally, extension
to large deformation and including plasticity into the damage formulation is of great interest [80].
Acknowledgements: Financial support of Subproject A6 of the Transregional Collaborative
Research Center SFB/TRR 87 and Subproject A01 of the Transregional Collaborative Research
Center SFB-TRR 280 by the German Research Foundation (DFG) is gratefully acknowledged.
6. Appendix A: Analytical solution for 1-D damage sub problem
In this appendix, a closed-form solution for 1-D damage PDE is presented. By which the
difference between standard and cohesive PF models, can be interpreted. Readers are also
encourage to see [35, 4]. To simplify the equations following function is introduced as:
g(d) =1
fD− 1⇒ g′(d) =
−f ′DfD
2 . (49)
Which reads:
ε(d) =σ
E0
fD−1 =
σ
E0
(g(d) + 1). (50)
With having damage PDE in one hand and the 1-D elastic energy as ψe,1D =1
2Eε2, the PDE
of damage can be rewritten as:
σ2g′(d)
2E0
− Gc
c0lc
(ω′(d)− 2l2cd,xx
)= 0. (51)
Assuming uniform damage, the latter term can be neglected.
σ2g′(d)
2E0
− Gc
c0lcω′(d) = 0 (52)
36
As a result, the following equations are obtained for strain and stress at the onset of crack
initiation (d = 0): σ =
√2E0Gc
c0lc
ω′(0)
g′(0)
ε =1
E0
√−2E0Gc
c0lc
ω′(0)
f ′Dc.
(53)
The above expressions are obtained, by using L’Hopital’s rule since the limit is indeterminate
(limd→0ω(d)
g(d)= 0
0).
A linear term in the crack topology function yields an initial elastic stage before damage
initiation, and the maximum stress is achieved, when d = 0. On the contrary and in the
standard phase-field approach, the damage initiates from infinitesimal tensile strain, and stress
reaches its maximum value when d = 0.25. Recalling Eq. 53, a1 computed as
σ =
√2E0Gc
c0lc
2
a1⇒ a1 =
2EGc
σ20c0lc
, (54)
which grantees the value of maximum stress to be σ0 independently of the internal length scale
lc. Considering the TSL (depicted in Fig. 5), lim[[u]]→λf σ([[u]]) = 0 is accomplished with having
the final crack opening as:
Wu =2πGc
σ0c0
√2(1 + a2). (55)
Having the c0 = π and λf =2Gc,0
σ0in hand, for fulfilling λf = Wu, we have
a2 = −0.5. (56)
7. Appendix B: Regularised crack density function
In this appendix we provide and review some information regarding a general form of the
crack density function introduced in Eqs. 10 and 22
γ(d,∇d) =1
c0
(ω(d)
lc+ lc∇d · ∇d
)(57)
According to Euler-Lagrange principle, the governing equation for phase-field damage is ob-
tained as dω(d)
dd− 2l2c ∆d = 0 in Ω
∇d · n = 0 on ∂Ω
(58)
37
By multiplying the above equation by d′ and integrating along the normal direction to the
crack direction one obtains:ω(d)− l2c |∇n|2 = 0⇒ γ =
2
c0lcω(d)
|∇n| :=∂d
∂|xn|=
1
lc
√ω(d)
(59)
Here, we defined xn := (x− xc) · nc, where xn is scalar product of the vector which obtained
as the distance of point x from its closest point at the surface of crack xc. The normal vector
to the crack surface is denoted by nc. Considering dV = 2|dxn| · As and Eq.22 reads:
Γc =
∫B
γdV =4
c0
∫ d
0
ω(d)1
lcd|xn| · As (60)
where As is the surface of crack. Finally, it follows as:
Γc = As ⇒ c0 = 4
∫ d
0
ω(d)1
lcd|xn| = 4
∫ 1
0
√ω(β)dβ (61)
Different groups of function can be chosen for ω(d), nevertheless they should fulfill the following
conditions: ω(0) = 0, ω(1) = 1, ω′ ≥ 0. Some choices for crack geometric function in PF
damage models are [9, 81, 82, 35]:
ω(d) = d2
du(x) = exp
(−|x|lc
)Du = +∞
c0 = 2
,
ω(d) = d
du(x) =
(1− −|x|
2lc
)2
Du = 2lc
c0 =8
3
,
ω(d) = 2d− d2
du(x) = 1− sin
(|x|lc
)Du =
π
2lc
c0 = π
(62)
Note that by having the linear term in the crack topology function, one can introduce the
threshold for damage. In other words, for the first choice (ω(d) = d2), damage zone expands
towards infinity. Where, Du denotes to the damage half bandwidth.
8. Appendix C: Derivation of anisotropic crack energy using structural tensor
The expression in Eq. 13 for the crack energy using the structural tensor A is derived in
this appendix. Note that c(•) and s(•) denote the functions cos(•) and sin(•), respectively.
38
Recalling Eq. 9 and the definition of A in Eq. 12 we have the following equation for the fracture
energy.
ψc,a =Gc,0
c0lc
(ω(d) + l2c ∇dT A∇d
)(63)
=Gc,0
c0lc
ω(d) + αl2c ∇dT c2(φ) c(φ)s(φ)
c(φ)s(φ) s2(φ)
||∇d||2∇d + l2c ∇dT ∇d
. (64)
Considering Eq. 8, the direction of ∇d is denoted by the angle β = atan
(∇d · e2∇d · e1
)= θ + π/2.
Therefore, we have ∇dT =[c(β) s(β)
]. One can further simplify the above expression as
ψc,a =Gc,0
c0lc
ω(d) + αl2c
[c(β) s(β)
]c2(φ) c(β) + c(φ) c(β) s(β)
c(φ) s(φ) c(β) + s2(φ) s(β)
||∇d||2 + l2c ∇dT ∇d
,
(65)
=Gc,0
c0lc
(ω(d) + αl2c (c(φ)c(β) + s(φ)s(β))2 + l2c∇dT∇d
), (66)
=Gc,0
c0lc
(ω(d) + αl2c cos2(β − φ)||∇d||2 + l2c∇dT∇d
). (67)
By reconsidering β = θ + π/2, we have:
ψc,a =Gc,0
c0lc
(ω(d) +
(1 + αl2c sin2(θ − φ)
)||∇d||2
). (68)
39
References
[1] S. Rezaei, J. R. Mianroodi, T. Brepols, S. Reese, Direction-dependent fracture in solids:
Atomistically calibrated phase-field and cohesive zone model, Journal of the Mechanics and
Physics of Solids 147 (2021) 104253.
[2] E. Lorentz, S. Cuvilliez, K. Kazymyrenko, Convergence of a gradient damage model toward
a cohesive zone model, Comptes Rendus Mecanique 339 (1) (2011) 20–26.
[3] J.-Y. Wu, V. P. Nguyen, A length scale insensitive phase-field damage model for brittle
fracture, Journal of the Mechanics and Physics of Solids 119 (2018) 20 – 42.
[4] R. J. Geelen, Y. Liu, T. Hu, M. R. Tupek, J. E. Dolbow, A phase-field formulation for
dynamic cohesive fracture, Computer Methods in Applied Mechanics and Engineering 348
(2019) 680–711.
[5] A. A. Griffith, Vi. the phenomena of rupture and flow in solids, Philosophical transactions
of the royal society of london. Series A, containing papers of a mathematical or physical
character 221 (582-593) (1921) 163–198.
[6] G. I. Barenblatt, The mathematical theory of equilibrium cracks in brittle fracture, Ad-
vances in applied mechanics 7 (1) (1962) 55–129.
[7] G. Francfort, J.-J. Marigo, Revisiting brittle fracture as an energy minimization problem,
Journal of the Mechanics and Physics of Solids 46 (8) (1998) 1319–1342.
[8] B. Bourdin, G. A. Francfort, J.-J. Marigo, Numerical experiments in revisited brittle frac-
ture, Journal of the Mechanics and Physics of Solids 48 (4) (2000) 797–826.
[9] C. Miehe, F. Welschinger, M. Hofacker, Thermodynamically consistent phase-field models
of fracture: Variational principles and multi-field fe implementations, International Journal
for Numerical Methods in Engineering 83 (10) (2010) 1273–1311.
[10] D. Schneider, E. Schoof, Y. Huang, M. Selzer, B. Nestler, Phase-field modeling of crack
propagation in multiphase systems, Computer Methods in Applied Mechanics and Engi-
neering 312 (2016) 186 – 195.
40
[11] B.-X. Xu, Y. Zhao, P. Stein, Phase field modeling of electrochemically induced fracture in
li-ion battery with large deformation and phase segregation, GAMM-Mitteilungen 39 (1)
(2016) 92–109.
[12] E. Martinez-Paneda, A. Golahmar, C. F. Niordson, A phase field formulation for hydrogen
assisted cracking, Computer Methods in Applied Mechanics and Engineering 342 (2018)
742–761.
[13] E. Moshkelgosha, M. Mamivand, Concurrent modeling of martensitic transformation and
crack growth in polycrystalline shape memory ceramics, Engineering Fracture Mechanics
241 (2021) 107403.
[14] T. Q. Bui, X. Hu, A review of phase-field models, fundamentals and their applications to
composite laminates, Engineering Fracture Mechanics 248 (2021) 107705.
[15] C. Steinke, I. Zreid, M. Kaliske, On the relation between phase-field crack approximation
and gradient damage modelling, Computational Mechanics 59 (2017).
[16] T. Linse, P. Hennig, M. Kastner, R. Borst, A convergence study of phase-field models for
brittle fracture, Engineering Fracture Mechanics 184 (2017).
[17] H. Amor, J.-J. Marigo, C. Maurini, Regularized formulation of the variational brittle frac-
ture with unilateral contact: Numerical experiments, Journal of the Mechanics and Physics
of Solids 57 (8) (2009) 1209 – 1229.
[18] C. Kuhn, R. Muller, Simulation of size effects by a phase field model for fracture, Theoret-
ical and Applied Mechanics Letters 4 (5) (2014) 051008.
[19] M. J. Borden, C. V. Verhoosel, M. A. Scott, T. J. Hughes, C. M. Landis, A phase-field
description of dynamic brittle fracture, Computer Methods in Applied Mechanics and En-
gineering 217-220 (2012) 77–95.
[20] Nguyen T. T., Yvonnet J., Bornert M., Chateau C., Sab K., Romani R., Le Roy R., On
the choice of parameters in the phase field method for simulating crack initiation with
experimental validation, International Journal of Fracture 197 (2) (2016) 213–226.
41
[21] X. Zhang, C. Vignes, S. Sloan, D. Sheng, Numerical evaluation of the phase-field model for
brittle fracture with emphasis on the length scale, Computational Mechanics 59 (2017).
[22] C. V. Verhoosel, R. de Borst, A phase-field model for cohesive fracture, International
Journal for Numerical Methods in Engineering 96 (1) (2013) 43–62.
[23] J. Mergheim, E. Kuhl, P. Steinmann, A finite element method for the computational mod-
elling of cohesive cracks, International Journal for Numerical Methods in Engineering 63 (2)
(2005) 276–289.
[24] S. Rezaei, S. Wulfinghoff, S. Reese, Prediction of fracture and damage in micro/nano
coating systems using cohesive zone elements, International Journal of Solids and Structures
121 (2017) 62 – 74.
[25] S. Rezaei, D. Jaworek, J. R. Mianroodi, S. Wulfinghoff, S. Reese, Atomistically motivated
interface model to account for coupled plasticity and damage at grain boundaries, Journal
of the Mechanics and Physics of Solids 124 (2019) 325 – 349.
[26] S. Rezaei, J. R. Mianroodi, K. Khaledi, S. Reese, A nonlocal method for modeling interfaces:
Numerical simulation of decohesion and sliding at grain boundaries, Computer Methods in
Applied Mechanics and Engineering 362 (2020) 112836.
[27] J. S. K. L. Gibson, S. Rezaei, H. Ruess, M. Hans, D. Music, S. Wulfinghoff, J. M. Schneider,
S. Reese, S. Korte-Kerzel, From quantum to continuum mechanics: studying the fracture
toughness of transition metal nitrides and oxynitrides, Materials Research Letters 6 (2)
(2018) 142–151.
[28] E. Tanne, T. Li, B. Bourdin, J.-J. Marigo, C. Maurini, Crack nucleation in variational
phase-field models of brittle fracture, Journal of the Mechanics and Physics of Solids 110
(2017).
[29] A. Kumar, B. Bourdin, G. A. Francfort, O. Lopez-Pamies, Revisiting nucleation in the
phase-field approach to brittle fracture, Journal of the Mechanics and Physics of Solids 142
(2020) 104027.
42
[30] G. Molnar, A. Doitrand, R. Estevez, A. Gravouil, Toughness or strength? regularization in
phase-field fracture explained by the coupled criterion, Theoretical and Applied Fracture
Mechanics 109 (2020) 102736.
[31] T. Nguyen, J. Yvonnet, Q.-Z. Zhu, M. Bornert, C. Chateau, A phase-field method for
computational modeling of interfacial damage interacting with crack propagation in realistic
microstructures obtained by microtomography, Comput. Methods Appl. Mech. Eng. 312
(2016) 567 – 595.
[32] P. Tarafder, S. Dan, S. Ghosh, Finite deformation cohesive zone phase field model for crack
propagation in multi-phase microstructures, Comput. Mech. 66 (2020) 723–743.
[33] Gradient damage models: Toward full-scale computations, Computer Methods in Applied
Mechanics and Engineering 200 (21) (2011) 1927 – 1944.
[34] M. R. Tupek, Cohesive phase-field fracture and a pde constrained optimization approach
to fracture inverse problems (2016).
[35] J.-Y. Wu, A unified phase-field theory for the mechanics of damage and quasi-brittle failure,
Journal of the Mechanics and Physics of Solids 103 (2017) 72 – 99.
[36] J. Fang, C. Wu, T. Rabczuk, C. Wu, G. Sun, Q. Li, Phase field fracture in elasto-plastic
solids: a length-scale insensitive model for quasi-brittle materials, Comput Mech (2020)
931–961.
[37] F. Freddi, F. Iurlano, Numerical insight of a variational smeared approach to cohesive
fracture, Journal of the Mechanics and Physics of Solids 98 (2017) 156 – 171.
[38] V. Hakim, A. Karma, Crack path prediction in anisotropic brittle materials, Phys. Rev.
Lett. 95 (2005) 235501.
[39] Y. Gao, Z. Liu, Q. Zeng, T. Wang, Z. Zhuang, K.-C. Hwang, Theoretical and numerical
prediction of crack path in the material with anisotropic fracture toughness, Engineering
Fracture Mechanics 180 (2017) 330 – 347.
[40] S. Teichtmeister, D. Kienle, F. Aldakheel, M.-A. Keip, Phase field modeling of fracture in
anisotropic brittle solids, International Journal of Non-Linear Mechanics (2017).
43
[41] B. Li, C. Maurini, Crack kinking in a variational phase-field model of brittle fracture with
strongly anisotropic surface energy, Journal of the Mechanics and Physics of Solids 125
(2019) 502 – 522.
[42] B. Li, C. Peco, D. Millan, I. Arias, M. Arroyo, Phase-field modeling and simulation of
fracture in brittle materials with strongly anisotropic surface energy, International Journal
for Numerical Methods in Engineering 102 (3-4) (2015) 711–727.
[43] E. G. Kakouris, S. P. Triantafyllou, Phase-field material point method for dynamic brit-
tle fracture with isotropic and anisotropic surface energy, Computer Methods in Applied
Mechanics and Engineering 357 (2019) 112503.
[44] T. T. Nguyen, J. Rethore, M.-C. Baietto, Phase field modelling of anisotropic crack prop-
agation, Eet. J. Mech. A-Solid 65 (2017) 279 – 288.
[45] N. Nguyen, J. Yvonnet, J. e. a. Rethore, Identification of fracture models based on phase
field for crack propagation in heterogeneous lattices in a context of non-separated scales,
Comput Mech 63 (2019) 1047–1068.
[46] J. Eggleston, G. McFadden, P. Voorhees, A phase-field model for highly anisotropic inter-
facial energy, Physica D: Nonlinear Phenomena 150 (1) (2001) 91 – 103.
[47] S. Zhang, Chemomechanical modeling of lithiation-induced failure in high-volume-change
electrode materials for lithium ion batteries, Computational Materials (2017).
[48] S. Brach, M. Hossain, B. Bourdin, K. Bhattacharya, Anisotropy of the effective toughness
of layered media, Journal of the Mechanics and Physics of Solids 131 (2019) 96 – 111.
[49] M. Z. Hossain, T. Ahmed, B. Silverman, M. S. Khawaja, J. Calderon, A. Rutten, S. Tse,
Anisotropic toughness and strength in graphene and its atomistic origin, Journal of the
Mechanics and Physics of Solids 110 (2018) 118–136.
[50] U. Pillai, S. P. Triantafyllou, Y. Essa, F. M. de la Escalera, An anisotropic cohesive phase
field model for quasi-brittle fractures in thin fibre-reinforced composites, Composite Struc-
tures 252 (2020) 112635.
44
[51] P. Zhang, W. Yao, X. Hu, X. Zhuang, Phase field modelling of progressive failure in com-
posites combined with cohesive element with an explicit scheme, Composite Structures
(2020) 113353.
[52] T. K. Mandal, V. P. Nguyen, J.-Y. Wu, A length scale insensitive anisotropic phase field
fracture model for hyperelastic composites, International Journal of Mechanical Sciences
188 (2020) 105941.
[53] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent crack
propagation: Robust algorithmic implementation based on operator splits, Computer
Methods in Applied Mechanics and Engineering 199 (45) (2010) 2765–2778.
[54] T.-T. Nguyen, J. Rethore, M.-C. Baietto, Phase field modelling of anisotropic crack prop-
agation, European Journal of Mechanics - A/Solids 65 (2017) 279–288.
[55] S. May, J. Vignollet, R. de Borst, A numerical assessment of phase-field models for brittle
and cohesive fracture: Γ -Convergence and stress oscillations, European Journal of Mechan-
ics - A/Solids 52 (2015) 72–84.
[56] C. Miehe, L.-M. Schanzel, H. Ulmer, Phase field modeling of fracture in multi-physics
problems. part i. balance of crack surface and failure criteria for brittle crack propagation
in thermo-elastic solids, Computer Methods in Applied Mechanics and Engineering 294
(2015) 449–485.
[57] B. Yin, M. Kaliske, An anisotropic phase-field model based on the equivalent crack surface
energy density at finite strain, Computer Methods in Applied Mechanics and Engineering
369 (2020) 113202.
[58] K. Khaledi, S. Rezaei, S. Wulfinghoff, S. Reese, Modeling of joining by plastic deformation
using a bonding interface finite element, International Journal of Solids and Structures
(2018).
[59] T. Brepols, S. Wulfinghoff, S. Reese, Gradient-extended two-surface damage-plasticity:
Micromorphic formulation and numerical aspects, International Journal of Plasticity 97
(2017) 64 – 106.
45
[60] R. de Borst, C. Verhoosel, Gradient damage vs phase-field approaches for fracture: Similar-
ities and differences, Computer Methods in Applied Mechanics and Engineering 312 (De-
cember) (2016) 78–94.
[61] T. Gerasimov, L. De Lorenzis, A line search assisted monolithic approach for phase-field
computing of brittle fracture, Computer Methods in Applied Mechanics and Engineering
312 (2016) 276–303.
[62] P. Zhang, X. Hu, X. Wang, W. Yao, An iteration scheme for phase field model for cohesive
fracture and its implementation in abaqus, Engineering Fracture Mechanics 204 (2018).
[63] S. Zhang, W. Jiang, M. R. Tonks, A new phase field fracture model for brittle materials that
accounts for elastic anisotropy, Computer Methods in Applied Mechanics and Engineering
358 (2020) 112643.
[64] N. Singh, C. Verhoosel, R. de Borst, E. van Brummelen, A fracture-controlled path-
following technique for phase-field modeling of brittle fracture, Finite Elements in Analysis
and Design 113 (2016) 14–29.
[65] F. Fei, J. Choo, Double-phase-field formulation for mixed-mode fracture in rocks, Computer
Methods in Applied Mechanics and Engineering 376 (2021) 113655.
[66] P. Shanthraj, B. Svendsen, L. Sharma, F. Roters, D. Raabe, Elasto-viscoplastic phase field
modelling of anisotropic cleavage fracture, Journal of the Mechanics and Physics of Solids
99 (2017) 19–34.
[67] E. C. Bryant, W. Sun, A mixed-mode phase field fracture model in anisotropic rocks
with consistent kinematics, Computer Methods in Applied Mechanics and Engineering 342
(2018) 561 – 584.
[68] E. Lorentz, A nonlocal damage model for plain concrete consistent with cohesive fracture,
International Journal of Fracture 207 (2) (2017) 123–159.
[69] M. Nejati, A. Aminzadeh, F. Amann, M. O. Saar, T. Driesner, Mode i fracture growth in
anisotropic rocks: Theory and experiment, International Journal of Solids and Structures
195 (2020) 74–90.
46
[70] B. Vowinckel, T. Fruhwirt, J. Maßmann, T. Nagel, M. Nest, D. Ptschke, C. Rolke, A. S.
Sattari, P. Schmidt, H. Steeb, K. Yoshioka, G. Ziefle, O. Kolditz, Model-Experiment-
Exercises (MEX), Springer International Publishing, Cham, 2021, pp. 97–192.
[71] C. Schreiber, T. Ettrich, C. Kuhn, R. Muller, A Phase Field Modeling Approach of Crack
Growth in Materials with Anisotropic Fracture Toughness, in: C. Garth, J. C. Aurich,
B. Linke, R. Muller, B. Ravani, G. H. Weber, B. Kirsch (Eds.), 2nd International Con-
ference of the DFG International Research Training Group 2057 – Physical Modeling for
Virtual Manufacturing (iPMVM 2020), Vol. 89 of Open Access Series in Informatics (OA-
SIcs), 2021, pp. 9:1–9:17.
[72] P. Zhang, X. Hu, T. Q. Bui, W. Yao, Phase field modeling of fracture in fiber reinforced
composite laminate, International Journal of Mechanical Sciences 161-162 (2019) 105008.
[73] T.-T. Nguyen, J. Rethore, J. Yvonnet, M.-C. Baietto, Multi-phase-field modeling of
anisotropic crack propagation for polycrystalline materials, Computational Mechanics
60 (2) (2017) 289–314.
[74] M. Paggi, J. Reinoso, Revisiting the problem of a crack impinging on an interface:a mod-
eling framework for the interaction between the phase field approach for brittle fracture
and the interface cohesive zone model, Computer Methods in Applied Mechanics and En-
gineering 321 (2017) 145 – 172.
[75] A. Lotfolahpour, M. Asle Zaeem, Effects of cleavage plane and material strength on fracture
of polycrystalline brittle materials: A phase-field modeling study, Computational Materials
Science 197 (2021) 110642.
[76] T. K. Mandal, V. P. Nguyen, J.-Y. Wu, Comparative study of phase-field damage models
for hydrogen assisted cracking, Theoretical and Applied Fracture Mechanics 111 (2021)
102840.
[77] S. Rezaei, M. Arghavani, S. Wulfinghoff, N. C. Kruppe, T. Brogelmann, S. Reese,
K. Bobzin, A novel approach for the prediction of deformation and fracture in hard coat-
ings: Comparison of numerical modeling and nanoindentation tests, Mechanics of Materials
117 (2018) 192–201.
47
[78] S. Reese, T. Brepols, M. Fassin, L. Poggenpohl, S. Wulfinghoff, Using structural tensors
for inelastic material modeling in the finite strain regime – A novel approach to anisotropic
damage, Journal of the Mechanics and Physics of Solids 146 (2021) 104174.
[79] B. Yin, M. Kaliske, A ductile phase-field model based on degrading the fracture toughness:
Theory and implementation at small strain, Computer Methods in Applied Mechanics and
Engineering 366 (2020) 113068.
[80] T. Brepols, S. Wulfinghoff, S. Reese, A gradient-extended two-surface damage-plasticity
model for large deformations, International Journal of Plasticity 129 (2020) 102635.
[81] K. Pham, H. Amor, J.-J. Marigo, C. Maurini, Gradient damage models and their use to
approximate brittle fracture, International Journal of Damage Mechanics 20 (4) (2011)
618–652.
[82] B. Bourdin, J.-J. Marigo, C. Maurini, P. Sicsic, Morphogenesis and propagation of complex
cracks induced by thermal shocks, Phys. Rev. Lett. 112 (2014) 014301.
48